Skip to main content
Dryad logo

Dataset for: Physics-informed neural networks with monotonicity constraints for Richardson-Richards equation: Estimation of constitutive relationships and soil water flux density from volumetric water content measurements by Toshiyuki Bandai and Teamrat A. Ghezzehei

Citation

Bandai, Toshiyuki; Ghezzehei, Teamrat (2022), Dataset for: Physics-informed neural networks with monotonicity constraints for Richardson-Richards equation: Estimation of constitutive relationships and soil water flux density from volumetric water content measurements by Toshiyuki Bandai and Teamrat A. Ghezzehei, Dryad, Dataset, https://doi.org/10.6071/M3T376

Abstract

Simulation of soil moisture dynamics is important for various fields, such as agriculture, hydrological modeling, and natural disasters. Such simulations are conducted by solving a partial differential equation (PDE) called the Richardson-Richards equation (RRE). Because the RRE is a highly non-linear PDE, we need to solve it numerically. Various numerical methods have been used to solve the RRE, such as the finite difference, finite element, and finite element volume method. In order for those numerical methods to produce correct solutions, they require precise information on initial and boundary conditions, as well as hydraulic properties of soils. In "Physics-informed neural networks with monotonicity constraints for Richardson-Richards equation: Estimation of constitutive relationships and soil water flux density from volumetric water content measurements by Toshiyuki Bandai and Teamrat A. Ghezzehei," we presented an alternative numerical method to infer hydraulic properties of soils from volumetric water content measurements withtout the initial and boudary conditions using physics-informed neural networks (PINNs). In this repository, we provide all the datasets that are needed to reproduce the analysis conducted in the paper.

Methods

The dataset was collected by computational simulations. The zip files (PINN_Monotonic.zip, NN_structure.zip, relative_errors.zip) were produced by running PINNs (physics informed neural networks) programs the authors developed. Please refer to Github page (https://github.com/ToshiyukiBandai/PINNs_RRE) for the PINNs programs. The other zip files (hydrus_run_files.zip and hydrus_nod_files.zip) were produced by running a software called HYDRUS-1D (https://www.pc-progress.com/en/Default.aspx?hydrus-1d).

Usage Notes

The dataset contains 5 zip files.

“PINN_Monotonic.zip”

This zip file contains 6 files that have the results from the PINNs with monotonicity constraints for 3 soils and 2 upper boundary conditions. These files are used in Section 4.3. In each file, there is “data.csv,” which summarizes predicted values and training data. Also, trained WRCs and HCFs are “lookup.csv.”  Other data can be produced from “data.csv.”

“NN_structure.zip”

This zip file contains Excel files that contain the results from the PINNs with and without monotonicity constraints with different neural network architecture. These Excel files are used to generate ”relative_errors” files.

“relative_errors.zip”

This zip file contains relative error data used in Section 4.1 and 4.2.

“hydrus_run_files.zip”

This zip file contains files that are useful to run HYDRUS-1D and process the output data. “hydrus_condition.xlsx” summarizes the HYDRUS-1D settings. “HydrusProcess.ipynb” is a Jupyter notebook file that can be used to convert the HYDRUS data into the “{soil_name}_nod.csv” file, which is used to feed the training and validation data to the PINNs.

hydrus_nod_files.zip”

This zip file contains synthetic data generated by HYDRUS-1D. Each file corresponds to each soil (sandy loam, loam, silt loam soil) and different upper boundary conditions (“ ": Scenario 1; “2”: Scenario 2). These files can be generated by “hydrus_run_files.zip.”